Jude Dialog requirements against PolyAI (frozen 2026-10-23)
Each PolyAI claim, the deterministic target that exceeds it, the pass or fail test, and the journeys that prove it. Source research: [RESEARCH.md](/research). Built by Claude-Jegudiel 12:3323102026-1.
| Id | PolyAI claim | PolyAI figure | Target (exceeds) | Pass or fail test | Journeys | Source |
| P01 | At 90% per-turn accuracy only 12% of 20-turn conversations succeed (0.90^20 = 0.12) | 12% conversation success | 100% of conversations end with every promised outcome verified against the ledger, at 20, 40 and 60 turns, with failures injected | Harness: every long-conversation run; fail if any conversation has an unverified outcome | [dlg-adv-long-20](/journeys/dlg-adv-long-20), [dlg-adv-long-40](/journeys/dlg-adv-long-40), [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://poly.ai/blog/your-ai-fails-production |
| P02 | A customer service conversation runs to 20 turns | 20 turns | Conversations measured at 20, 40 and 60 turns | Fail if a long run ends below its turn target | [dlg-adv-long-20](/journeys/dlg-adv-long-20), [dlg-adv-long-40](/journeys/dlg-adv-long-40), [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://poly.ai/blog/your-ai-fails-production |
| P03 | Failure compounds from turn 6 | turn 6 | A corrupted turn 6 changes no state (state hash before = after) and the conversation still verifies | Fail if the receipt state hash changes on the corrupted turn or the outcome is not verified | [dlg-adv-corrupt-t6](/journeys/dlg-adv-corrupt-t6), [dlg-adv-long-20](/journeys/dlg-adv-long-20), [dlg-adv-long-40](/journeys/dlg-adv-long-40), [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://poly.ai/blog/your-ai-fails-production |
| P04 | By turn 14 the model reasons from a broken thread | turn 14 | Backend timeout from turn 14: zero facts stated while it is down, then recovery by retry or hand-off with a reference, 100% | Fail on any fact on a failed-backend turn, or no recovery | [dlg-adv-timeout-t14](/journeys/dlg-adv-timeout-t14), [dlg-adv-timeout-persistent](/journeys/dlg-adv-timeout-persistent), [dlg-adv-long-20](/journeys/dlg-adv-long-20) | https://poly.ai/blog/your-ai-fails-production |
| P05 | Production means 200 simultaneous calls | 200 simultaneous | At least 400 simultaneous conversations through the public address, every one verified | Harness spike runs at 200 and 400; fail if any conversation is unverified | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency) | https://poly.ai/blog/your-ai-fails-production |
| P06 | Latency is 2 seconds per turn under load | 2 s per turn | p95 turn latency under 500 ms at 200 and at 400 simultaneous conversations, measured by the caller through HTTPS | Fail if p95 >= 500 ms in either spike | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency) | https://poly.ai/blog/your-ai-fails-production |
| P07 | Dead air reaches 40 seconds per conversation | 40 s | Total wait per 20-turn conversation under 5 s at 200 simultaneous | Sum of turn latencies per conversation; fail if any 20-turn conversation waits 5 s or more | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency), [dlg-adv-long-20](/journeys/dlg-adv-long-20) | https://poly.ai/blog/your-ai-fails-production |
| P08 | Multilingual slip: an English reply to a Spanish caller | no figure | Zero replies in a language other than the caller's, including a mid-call switch and retrieval returning another language | Every reply's language is detected from template vocabulary; fail on any mismatch | [dlg-adv-lang-switch](/journeys/dlg-adv-lang-switch), [dlg-adv-retrieval-lang](/journeys/dlg-adv-retrieval-lang), [dlg-fin-faq](/journeys/dlg-fin-faq), [dlg-health-faq](/journeys/dlg-health-faq), [dlg-hotel-faq](/journeys/dlg-hotel-faq), [dlg-insurance-faq](/journeys/dlg-insurance-faq) and 5 more | https://poly.ai/blog/your-ai-fails-production |
| P09 | Hallucination when a backend API fails | no figure | Zero fabricated answers under injected timeouts and errors: the data is said to be not reachable, then retried or handed over | Fail if any failed-backend turn carries a fact or a figure | [dlg-adv-timeout-t14](/journeys/dlg-adv-timeout-t14), [dlg-adv-timeout-persistent](/journeys/dlg-adv-timeout-persistent) | https://poly.ai/blog/your-ai-fails-production |
| P10 | Prompt layering causes contradictions and decay | no figure | Constraints are code, not prompt: a payment asked at turn 59 without verification is refused; policy at turn 60 equals policy at turn 1 | Fail if the turn-59 payment request is not refused | [dlg-adv-policy-depth](/journeys/dlg-adv-policy-depth), [dlg-adv-auth-lockout](/journeys/dlg-adv-auth-lockout), [dlg-adv-sca-binding](/journeys/dlg-adv-sca-binding) | https://poly.ai/blog/your-ai-fails-production |
| P11 | Raven is fine-tuned on hundreds of thousands of production calls | SFT corpus | No trained generation in the reply path: every reply is a frozen template filled only from backend records | Every reply matches the frozen template set; fail on any other text | [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://poly.ai/blog/your-ai-fails-production |
| P12 | Three post-training stages (SFT, GRPO/DPO fusion, auto-reasoning) | 3 stages | Every change to what Jude understands passes the full regression suite before it is live; harmful changes are rejected | Gate result stored with hash; fail if a harmful change is accepted or a safe one refused | [dlg-adv-self-improve](/journeys/dlg-adv-self-improve) | https://poly.ai/blog/your-ai-fails-production |
| P13 | Auto-reasoning in fewer than 40 tokens | < 40 tokens | 0 model tokens per turn; every decision is a named act with its sources in a signed receipt | Fail if a turn has no receipt or its facts lack a source | [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://docs.poly.ai/beh%61vior/models/raven |
| P14 | Raven 3.5 reasons on about half of turns | about 50% of turns | 100% of turns explained: act, facts and sources recorded and chained | Receipt chain verifies end to end for every conversation | [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://poly.ai/blog/PolyAI-Raven-35-beats-GPT5 |
| P15 | Raven median latency under 300 ms | < 300 ms median | Median turn latency under 150 ms at 200 simultaneous conversations | Fail if p50 >= 150 ms | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency) | https://docs.poly.ai/beh%61vior/models/raven |
| P16 | Raven lists 24 languages | 24 languages | 30 languages served from frozen, hashed language packs | Count of languages with a complete pack; fail below 30 | [dlg-adv-lang-switch](/journeys/dlg-adv-lang-switch) | https://docs.poly.ai/beh%61vior/models/raven |
| P17 | PolyAI recommends median under 500 ms and p95 under 1 s | p95 < 1 s | p95 under 500 ms at 200 and 400 simultaneous | Fail if p95 >= 500 ms | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency) | https://poly.ai/blog/voice-ai-latency-customer-satisfaction |
| P18 | API price US$0.02 per connected minute | US$0.02 per minute | No per-minute fee; compute measured per turn | CPU milliseconds per turn reported from the service | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency) | https://poly.ai/pricing |
| P19 | Token prices US$2.00 / 0.20 / 5.00 per million | per-token billing | 0 model tokens in the reply path | Fail if any reply is generated outside the template set | [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://poly.ai/pricing |
| P20 | API concurrency 2 calls, or 10 with a card | 2 or 10 | No concurrency cap; 400 simultaneous measured | Spike at 400 completes with every conversation verified | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency) | https://poly.ai/pricing |
| P21 | ISO 27001 stated; SOC 2, PCI DSS, HIPAA, GDPR shown as logos | certifications | Every turn signed and hash-chained, verifiable by anyone at /api/dialog/receipts/<session> | Chain verification over every harness conversation | [dlg-adv-long-60](/journeys/dlg-adv-long-60) | https://poly.ai/security |
| P22 | Customer outcomes: PG&E, Carter's, SafeRide, UniCredit, Fogo de Chão, Audibel | case studies | First-customer use receipt: a completed journey run as X-X.TECH for the first customer, in the simulated regulated context | Use receipt present in every journey's level 9 record | [dlg-fin-billing](/journeys/dlg-fin-billing) | https://poly.ai/case-studies |
| P28 | Agent Studio trial for two months | 60 days | Open now: chat and voice at https://dialog.judiverse.tech with no sign-up | The page answers live | [dlg-adv-concurrency](/journeys/dlg-adv-concurrency) | https://poly.ai/blog/your-ai-fails-production |
Regulation enforced in code
| Rule | Gate | Journey |
| EU AI Act art 50(1) | The first turn of every conversation discloses an AI agent | every journey |
| EU AI Act art 50(2) | Every reply carries machine-readable ai_generated marking | every journey |
| PSRs 2017 reg 100; SCA RTS arts 4-5 | Two factors before account data; payment code bound to amount and payee | [dlg-adv-sca-binding](/journeys/dlg-adv-sca-binding) |
| UK GDPR arts 22A-22D; EU AI Act Annex III 5(b)-(c) | Credit, limit and claim decisions always go to a person | [dlg-adv-human-review](/journeys/dlg-adv-human-review) |
| FCA FG21/1; PRIN 2A.2; Ofcom GC C5 | Vulnerability words record extra support and offer a person | [dlg-adv-vulnerable](/journeys/dlg-adv-vulnerable) |
| NHS DTAC 2.0 (DCB0129/DCB0160) | Clinical red flags stop booking, direct to 999 and alert a clinician | [dlg-adv-redflag](/journeys/dlg-adv-redflag) |
| Ofcom GC C4; FCA DISP 1.6 | Complaints recorded at once with an eight-week final response date and the ombudsman route | [dlg-adv-complaint](/journeys/dlg-adv-complaint) |
| FCA PRIN 2A.6 | Cancel takes no more turns than a status check | dlg-*-orders |
| Equality Act 2010 s20 | Every use case reachable by voice and by chat | every industry journey |
| Welsh Language Act 1993 | Welsh served with the same templates and gates as English | Welsh journeys |
Use cases and industries
PolyAI's eight use cases across its nine industries are 72 journeys (dlg-<industry>-<use case>), each linked to the closest existing Bank, Clinic and Calls journeys rather than duplicating them. Nineteen adversarial journeys cover interruptions, a corrupted turn 6, an API timeout from turn 14, language switches, a 200-conversation spike, duplicate requests, complaints, self-improvement and the regulatory gates.